基于功率信号分析的光伏电站故障诊断方法

Fault Diagnosis Algorithm for PV Power Plant Based on Power Signal Analysis

为提高光伏电站故障诊断精度,提出一种基于功率信号分析的光伏电站故障诊断方法。首先,用卷积神经网络结合长短记忆CNN-LSTM(convolutional neural networks-long short-term memory)模型和岭回归模型对历史发电的时序信息进行充分挖掘,再依据实际与预测发电功率之间的动态时间规整DTW(dynamic time warping)距离进行电站故障检测;其次,提出一个基于实际发电功率频域特征的故障分类指标,建立分类规则库,将电站故障分为通信故障、设备故障、限电故障,结合故障影响等效发电小时数评估电站故障程度;最后,通过算例分析验证了该算法的有效性。

To improve the fault diagnosis accuracy of a PV power plant fault,a fault diagnosis method for PV power plant based on power signal analysis is proposed.First,a convolutional neural networks-long short-term memory(CNNLSTM)network model and a ridge regression model are used to mine the time series information about the historical power generation data,and the dynamic time warping(DTW)distance between the actual and predicted power generation is selected to detect fault.Second,a fault classification index based on the frequency-domain characteristics of actual power generation is put forward,and the classification rules are built to classify the power plant faults into communication fault,equipment fault and power cut fault,and the fault impact equivalent power generation hours are combined to assess the degree of each type of fault.Finally,the analysis of an example verifies the effectiveness of the proposed algorithm.